MCP Connectors and the MCP server are Beta features for environments with UnqorkAI. Test each connector and tool, and verify the results before relying on them.
The Model Context Protocol (MCP) is an open standard that lets AI agents work with other systems. Unqork uses MCP in two directions. With MCP Connectors, the Build Agent reaches out to systems like Jira and Google Drive. With the MCP server, AI agents outside Unqork call API modules in an Unqork environment.
This article explains what MCP is, how the two directions differ, and which articles cover each one.
What MCP Is
An AI agent is software that uses an AI model to decide what to do next. By itself, an agent can only work with the information it receives. To look something up or make a change in another system, the agent needs a tool. A tool is one action a system offers, like "search Jira issues" or "look up a policy."
MCP gives agents one common way to find and use those tools. The parts of MCP include:
MCP server: The part of a system that lists its tools and runs a tool when an agent asks.
MCP client: The part of an agent that connects to an MCP server.
Tool: One action an MCP server offers. Each tool has a name, a description, the inputs it needs, and the outputs it returns.
Because every system follows the same standard, one agent can work with many systems. A system only has to offer an MCP server once for any agent to use it.
To learn more about the standard, see the Model Context Protocol website.
Two Ways Unqork Uses MCP
Unqork acts as an MCP client in one feature and as an MCP server in the other. The following table compares them.
Topic | MCP Connectors | MCP Server |
|---|---|---|
Direction | The Build Agent calls out to outside systems. | Agents outside Unqork call in to Unqork. |
Unqork's role | MCP client. | MCP server. |
What it connects | A list of vetted systems that Unqork maintains: Atlassian, Google Drive, Google Docs, Exa, and Firecrawl. | API modules that a Creator publishes as tools. |
Who uses it | Creators, while they build in the Module Builder. | AI agents that a company runs outside Unqork. |
Who sets it up | An administrator adds each connector. Each Creator then connects their own account. | A Creator publishes each module as a tool. An administrator creates the agent's API credential. |
How access works | The Build Agent reaches only what the Creator can reach in the outside system. | The agent's Express role decides which tools it can see and call. |
Where to set it up | The MCP Connectors administration page. See MCP Connectors. | The module's Settings dialog, on the MCP Settings page. |
MCP Connectors
MCP Connectors let the Build Agent read from and write to systems outside Unqork. For example, a Creator can ask the Build Agent to read a Jira issue and build a module from its requirements. The Build Agent can also summarize a Google Doc or search the web for an API reference.
An administrator adds each connector on the MCP Connectors Administration page, from a list that Unqork tests and maintains. To open the page, navigate to 
Administration > 
Environment Settings > 
Build Agent > MCP Connectors. Creators then connect from the Build Agent chat and choose whether each read or write needs their approval. Content a connector retrieves is sent to the Build Agent's model.
For the available connectors, how Creators sign in, and who does what, see MCP Connectors for the Build Agent.
MCP Server
The MCP server lets AI agents that run outside Unqork call API modules as tools. For example, a company's customer support agent could call an API module that looks up the status of a claim.
Each environment has one MCP endpoint. A Creator chooses which API modules to publish there, and the Build Agent can write the first tool setup. An agent signs in with an Express API credential, and every call is rate-limited and logged.
For how the MCP server works, see MCP Server for API Modules. To publish a tool, see Publish API Modules as MCP Tools.
Using Both Features
The two features are separate. Setting up one does not set up the other, and an environment can use either one or both. A connector gives the Build Agent tools from outside systems. A published tool lets outside agents use Unqork. Connectors work only with the Build Agent, and the MCP server publishes only API modules.
MCP Articles
The following articles cover each feature in detail.
MCP Connectors Articles
Article | Description |
|---|---|
What connectors are, the available connectors, sign-in types, and read and write tools. | |
Add, share, and manage connectors, and choose which tools the Build Agent can use. | |
Create the Google project and OAuth client that the Google connectors need. | |
Connect an account from the Build Agent chat and choose Ask or Trusted. |
MCP Server Articles
Article | Description |
|---|---|
How the MCP server works, including access, tokens, limits, responses, and logging. | |
Publish an API module as a tool and give an AI agent access to it. | |
Every setting on the module page where Creators publish and edit a tool. | |
Create the Express API credential an agent signs in with. | |
Set the Express role that decides which tools an agent can call. |
Related Articles
Article | Description |
|---|---|
What the Build Agent can and cannot do, including its MCP capabilities. | |
How prompt and connector content reach the model provider. |
Changelog
Date | Change |
|---|---|
2026-11-12 | Linked MCP Connectors for the Build Agent, MCP Server for API Modules, and MCP Settings in the article body and See Also. |
2026-11-09 | Added the navigation with icons to the MCP Connectors section. Glossary tooltips now cover MCP, open standard, and AI agent. |
2026-11-06 | Created the article to explain MCP and link the MCP Connectors and MCP server articles. |